Aaron Ballman

> Do we want these warnings generated by default? I think it should be opt-in Why do you think they should be opt-in as opposed to opt-out?

> > Do we want these warnings generated by default? I think it should be opt-in > > Why do you think they should be opt-in as opposed to opt-out?...

> @AaronBallman > > > My intuition is that users would want to know when their floating-point calculations are going to magically change their precision to be different than what...

> > FWIW, I'm skeptical that off-by-default warnings are worth the maintenance burdens. Community experience was that only a very small percentage of people ever opt in to an off-by-default...

> > It _might_ be reasonable to have an on-by-default diagnostic for double literals and maybe a related off-by-default diagnostic for long double literals and direct use of the types....

> @AaronBallman > > > Have you followed up to see if this suggestion works or not? (I'd slightly prefer the changes in one patch rather than a series of...

> > I believe only if the emitted target device code has operations with double LLVM type, there must be a warning. But it is not clear to me which...

I think I figured out what bothers me about this approach and why I keep getting confused. SYCL isn't doing anything odd here, there's nothing SYCL-specific about it at all....

> > The issue is that the user needs to know when they're getting conversions from float to double that they didn't expect to get. > > Sorry, I am...

> @intel/dpcpp-cfe-reviewers I am in the process of upstreaming this change. Can this PR be merged if approved or does it have to wait until its upstreamed? Also, [Linux /...